P-wave arrival picking using Chinese strong-motion acceleration records based on PhaseNet
Rapid and accurate P-wave arrival picking is one of the basic works of earthquake early warning.U-shaped networks such as PhaseNet have achieved good P/S arrival picking performance on the"DiTing"Chinese seismic velocity data,mainly focus on the earthquakes with ML less than 3.0.Currently,there are few studies on Chinese strong-motion records which belong to small data volume.The earthquake early warning(EEW)mainly focus on the earthquakes with ML greater than 3.0.For quickly and accurately picking up P-arrival on EEW,we explore the PhaseNets whether could achieve good performance which are trained using the small-data-volume Chinese strong-motion acceleration records As a result of the limitation of the longer time window of the model's inputs and the difference between the velocity and acceleration records,it is not totally suitable for the P-wave arrival picking when using the Chinese strong-motion acceleration records.To pick the P-wave arrivals under short time window in earthquake early warning accurately and rapidly,we rebuild and transfer the PhaseNet and its derivative networks using the China strong-motion acceleration records.The results show that the precision,recall,F1 score,picking error mean(μ),and standard deviation(δ)of PhaseNet and its derived network models are approximately 0.942、0.930、0.937,-20(ms),and 200(ms).These models have accurate and generalization performance on P-wave arrival.In addition,PhaseNet and its derivative networks perform well on high signal-to-noise records,and still need to be improved on the low signal-to-noise ratio records.
deep learningChinese strong-motion acceleration recordsP-wave arrival pickingPhaseNetearthquake early warning
万方数据
维普
